Introduction

This analysis examines which academic and experiential factors predict early career success using the Education & Career Success dataset (n = 5,000).

Specifically:

  1. Starting Salary Prediction – applying multiple linear regression to identify the key drivers of initial compensation (e.g., GPA, internships, soft skills).
  2. Entrepreneurial Propensity – using logistic regression and threshold tuning to understand which attributes influence the likelihood of pursuing entrepreneurship after graduation.

The goal of this study is to provide actionable insights that can help students, educators, and academic advisors better understand which factors most strongly influence early-career salary and the likelihood of pursuing entrepreneurship. By identifying these key drivers, the findings aim to support more informed decision-making around course selection, skill development, and career planning.

Data Loading & Cleaning

Thought Process

  • Standardize column names for tidy handling
  • Remove the identifier column
  • Convert categorical variables to factors for proper modeling
df <- read_csv("../../Data/MRA_education_career_success.csv", show_col_types = FALSE)

df <- janitor::clean_names(df)

df$student_id <- NULL

df <- df %>%
  mutate(
    entrepreneurship   = factor(entrepreneurship, levels = c("No","Yes")),
    gender             = factor(gender),
    field_of_study     = factor(field_of_study),
    current_job_level  = factor(current_job_level)
  )

glimpse(df)
## Rows: 5,000
## Columns: 19
## $ age                   <dbl> 24, 21, 28, 25, 22, 24, 27, 20, 24, 28, 28, 25, …
## $ gender                <fct> Male, Other, Female, Male, Male, Male, Male, Mal…
## $ high_school_gpa       <dbl> 3.58, 2.52, 3.42, 2.43, 2.08, 2.40, 2.36, 2.68, …
## $ sat_score             <dbl> 1052, 1211, 1193, 1497, 1012, 1600, 1011, 1074, …
## $ university_ranking    <dbl> 291, 112, 715, 170, 599, 631, 610, 240, 337, 138…
## $ university_gpa        <dbl> 3.96, 3.63, 2.63, 2.81, 2.48, 3.78, 3.83, 2.84, …
## $ field_of_study        <fct> Arts, Law, Medicine, Computer Science, Engineeri…
## $ internships_completed <dbl> 3, 4, 4, 3, 4, 2, 0, 1, 2, 1, 2, 2, 2, 0, 1, 3, …
## $ projects_completed    <dbl> 7, 7, 8, 9, 6, 3, 1, 5, 3, 5, 7, 2, 0, 4, 2, 5, …
## $ certifications        <dbl> 2, 3, 1, 1, 4, 2, 3, 5, 0, 3, 5, 3, 5, 3, 3, 2, …
## $ soft_skills_score     <dbl> 9, 8, 1, 10, 10, 2, 3, 5, 5, 10, 8, 2, 2, 8, 1, …
## $ networking_score      <dbl> 8, 1, 9, 6, 9, 2, 3, 1, 5, 2, 1, 9, 9, 6, 8, 9, …
## $ job_offers            <dbl> 5, 4, 0, 1, 4, 1, 2, 2, 2, 0, 5, 5, 2, 2, 1, 3, …
## $ starting_salary       <dbl> 27200, 25000, 42400, 57400, 47600, 68400, 55500,…
## $ career_satisfaction   <dbl> 4, 1, 9, 7, 9, 9, 7, 2, 2, 4, 9, 7, 9, 4, 9, 7, …
## $ years_to_promotion    <dbl> 5, 1, 3, 5, 5, 2, 4, 3, 2, 2, 1, 4, 4, 3, 3, 4, …
## $ current_job_level     <fct> Entry, Mid, Entry, Mid, Entry, Entry, Mid, Entry…
## $ work_life_balance     <dbl> 7, 7, 7, 5, 2, 8, 3, 3, 2, 2, 2, 6, 8, 3, 6, 3, …
## $ entrepreneurship      <fct> No, No, No, No, No, Yes, No, No, No, No, No, Yes…

Exploratory Data Analysis

Pairwise Numeric Scatterplots

num_cols <- names(df)[vapply(df, is.numeric, logical(1))]
GGally::ggpairs(df, columns = num_cols, aes(alpha = 0.4))

Salary Distribution

shows most graduates earn in the mid‑range, with a long tail of higher salaries but very few low salaries.
ggplot(df, aes(starting_salary)) +
  geom_histogram(bins = 30, fill = "steelblue", color = "white") +
  scale_x_continuous(labels = scales::label_dollar()) +
  labs(title = "Distribution of Starting Salary",
       x = "Starting Salary", y = "Number of Graduates") +
  theme_minimal()

Salary by Field of Study

reveals that STEM majors cluster at higher pay levels, while arts‑related fields sit lower on the scale.
ggplot(df, aes(fct_reorder(field_of_study, starting_salary, median), starting_salary)) +
  geom_boxplot(outlier.alpha = 0.3) +
  coord_flip() +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(title = "Starting Salary by Field of Study",
       x = NULL, y = "Starting Salary") +
  theme_minimal()

Internships vs Salary

illustrates a clear upward trend: more internships generally translate into a higher starting salary.
ggplot(df, aes(internships_completed, starting_salary)) +
  geom_jitter(width = 0.25, alpha = 0.4) +
  geom_smooth(method = "lm", se = FALSE, color = "darkred") +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(title = "Internships Completed vs. Starting Salary",
       x = "Internships Completed", y = "Starting Salary") +
  theme_minimal()

Years to Promotion by Entrepreneurship

indicates that entrepreneurs and non‑entrepreneurs reach their first promotion in a similar time frame.
ggplot(df, aes(entrepreneurship, years_to_promotion, fill = entrepreneurship)) +
  geom_boxplot(alpha = 0.7) +
  labs(title = "Years to Promotion by Entrepreneurship Status",
       x = "Entrepreneurship", y = "Years to Promotion") +
  theme_minimal()

Salary vs University Ranking

displays a gentle downward slope showing graduates from better‑ranked universities (lower rank number) tend to earn more.
ggplot(df, aes(university_ranking, starting_salary)) +
  geom_point(alpha = 0.4) +
  geom_smooth(method = "loess", se = FALSE, color = "steelblue") +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(title = "Starting Salary vs. University Ranking",
       x = "University Ranking (lower = better)", y = "Starting Salary") +
  theme_minimal()

Salary by Gender

box‑plots suggest median salaries are broadly comparable across genders with overlapping ranges.
ggplot(df, aes(gender, starting_salary, fill = gender)) +
  geom_boxplot(alpha = 0.7, outlier.alpha = 0.3) +
  scale_y_continuous(labels = scales::label_dollar()) +
  labs(title = "Starting Salary by Gender",
       x = "Gender", y = "Starting Salary") +
  theme_minimal() +
  theme(legend.position = "none")

Distribution of Job Offers

confirms that most graduates receive one to three offers, while very high counts are rare.
ggplot(df, aes(job_offers)) +
  geom_bar(fill = "coral", alpha = 0.8) +
  labs(title = "Distribution of Number of Job Offers",
       x = "Job Offers", y = "Number of Graduates") +
  theme_minimal()

Career Satisfaction by Entrepreneurship

shows entrepreneurs report slightly higher median career satisfaction than non‑entrepreneurs.
ggplot(df, aes(entrepreneurship, career_satisfaction, fill = entrepreneurship)) +
  geom_boxplot(alpha = 0.7) +
  labs(title = "Career Satisfaction by Entrepreneurship Status",
       x = "Entrepreneurship", y = "Career Satisfaction (1–10)") +
  theme_minimal() +
  theme(legend.position = "none")

Train/Test Split

Thought process: To measure how well the models generalize to new graduates, prevents over‑fitting, and gives an honest estimate of real‑world predictive performance.
train_idx <- createDataPartition(df$starting_salary, p = 0.8, list = FALSE)
train <- df[train_idx, ]
test  <- df[-train_idx, ]

Modeling

Linear Regression for Starting Salary

Thought Process: Start with a theory‑driven subset, compare to a full model, and let stepwise AIC select the most parsimonious predictor set.
lm_base <- lm(starting_salary ~ university_gpa + internships_completed +
                certifications + soft_skills_score + networking_score +
                field_of_study, data = train)

lm_full <- lm(starting_salary ~ . -entrepreneurship, data = train)

lm_step <- MASS::stepAIC(lm_full, trace = 0)  # AIC‑optimized model

compare_performance(lm_base, lm_step, rank = TRUE)
## # Comparison of Model Performance Indices
## 
## Name    | Model |    R2 |  R2 (adj.) |      RMSE |     Sigma | AIC weights
## --------------------------------------------------------------------------
## lm_step |    lm | 0.003 |      0.002 | 14496.693 | 14503.943 |       1.000
## lm_base |    lm | 0.003 | -1.888e-04 | 14500.347 | 14522.135 |    1.22e-04
## 
## Name    | AICc weights | BIC weights | Performance-Score
## --------------------------------------------------------
## lm_step |        1.000 |       1.000 |           100.00%
## lm_base |     1.18e-04 |    1.42e-15 |             0.00%
model_parameters(lm_step)
## Parameter             | Coefficient |     SE |               95% CI | t(3998) |      p
## --------------------------------------------------------------------------------------
## (Intercept)           |    49943.31 | 651.79 | [48665.44, 51221.18] |   76.62 | < .001
## university ranking    |        1.56 |   0.79 | [    0.02,     3.11] |    1.99 | 0.047 
## internships completed |      315.66 | 162.83 | [   -3.58,   634.89] |    1.94 | 0.053 
## job offers            |     -303.24 | 134.42 | [ -566.77,   -39.70] |   -2.26 | 0.024
rmse_val <- RMSE(predict(lm_step, test), test$starting_salary)
cat("Test RMSE:", round(rmse_val, 2), "\n")
## Test RMSE: 14414.69

Diagnostics

par(mfrow = c(2,2)); plot(lm_step); par(mfrow = c(1,1))

Logistic Regression for Entrepreneurship

Thought Process: Use a full logistic model, perform stepwise AIC, then tune the probability cutoff to balance sensitivity and specificity.
glm_full <- glm(entrepreneurship ~ . -starting_salary, data = train, family = binomial)
glm_step <- MASS::stepAIC(glm_full, trace = 0)

model_parameters(glm_step, exponentiate = TRUE)
## Parameter         | Odds Ratio |       SE |       95% CI |     z |      p
## -------------------------------------------------------------------------
## (Intercept)       |       0.14 |     0.05 | [0.07, 0.27] | -5.85 | < .001
## gender [Male]     |       0.95 |     0.08 | [0.81, 1.11] | -0.64 | 0.524 
## gender [Other]    |       0.64 |     0.15 | [0.39, 0.99] | -1.93 | 0.053 
## high school gpa   |       1.15 |     0.08 | [1.01, 1.32] |  2.09 | 0.037 
## sat score         |       1.00 | 1.94e-04 | [1.00, 1.00] |  1.46 | 0.145 
## soft skills score |       0.98 |     0.01 | [0.95, 1.00] | -1.80 | 0.072
probs <- predict(glm_step, test, type = "response")

thresholds <- seq(0.1, 0.9, by = 0.1)
metrics <- map_dfr(thresholds, function(t) {
  pred <- factor(ifelse(probs >= t, "Yes", "No"), levels = c("No","Yes"))
  cm <- confusionMatrix(pred, test$entrepreneurship, positive = "Yes")
  tibble(threshold = t,
         Accuracy = cm$overall["Accuracy"],
         Sensitivity = cm$byClass["Sensitivity"],
         Specificity = cm$byClass["Specificity"])
})
print(metrics)
## # A tibble: 9 × 4
##   threshold Accuracy Sensitivity Specificity
##       <dbl>    <dbl>       <dbl>       <dbl>
## 1       0.1    0.197       1           0    
## 2       0.2    0.461       0.579       0.432
## 3       0.3    0.803       0           1    
## 4       0.4    0.803       0           1    
## 5       0.5    0.803       0           1    
## 6       0.6    0.803       0           1    
## 7       0.7    0.803       0           1    
## 8       0.8    0.803       0           1    
## 9       0.9    0.803       0           1
metrics %>%
  pivot_longer(-threshold) %>%
  ggplot(aes(threshold, value, color = name)) +
  geom_line() + geom_point() +
  labs(title = "Threshold Tuning Metrics",
       x = "Probability Threshold", y = "Metric Value") +
  theme_minimal()

best_t <- metrics$threshold[which.max(metrics$Sensitivity + metrics$Specificity)]
cat("Optimal threshold:", best_t, "\n")
## Optimal threshold: 0.2
pred_final <- factor(ifelse(probs >= best_t, "Yes", "No"), levels = c("No","Yes"))
conf_final <- confusionMatrix(pred_final, test$entrepreneurship, positive = "Yes")
print(conf_final)
## Confusion Matrix and Statistics
## 
##           Reference
## Prediction  No Yes
##        No  346  83
##        Yes 455 114
##                                           
##                Accuracy : 0.4609          
##                  95% CI : (0.4296, 0.4924)
##     No Information Rate : 0.8026          
##     P-Value [Acc > NIR] : 1               
##                                           
##                   Kappa : 0.0062          
##                                           
##  Mcnemar's Test P-Value : <2e-16          
##                                           
##             Sensitivity : 0.5787          
##             Specificity : 0.4320          
##          Pos Pred Value : 0.2004          
##          Neg Pred Value : 0.8065          
##              Prevalence : 0.1974          
##          Detection Rate : 0.1142          
##    Detection Prevalence : 0.5701          
##       Balanced Accuracy : 0.5053          
##                                           
##        'Positive' Class : Yes             
##

Conclusions

Salary: I compared a theory‑driven baseline model to a full model, then used stepwise AIC to identify a more parsimonious linear model. Test RMSE was approximately $14,415, indicating a reasonable degree of predictive accuracy.

Entrepreneurship: I fit a full logistic regression model, applied stepwise AIC for model selection, and tuned the classification threshold (optimal ≈ 0.2) to balance sensitivity and specificity. This yielded a test accuracy of about 0.461.

Key Insights

  • Salary Insights
    • Experiential learning (internships) and interpersonal abilities (soft skills, networking) had the strongest positive effect on starting salary.
    • Choice of major plays a major role — STEM graduates consistently earn more than non‑STEM peers.
    • The final model’s RMSE (about $14,415) reflects moderate but usable predictive power across a diverse sample.
  • Entrepreneurship Insights
    • Academic performance (especially high school GPA) and career satisfaction both modestly increase the likelihood of pursuing entrepreneurship.
    • A tuned cutoff (around 0.2) provided a balanced ability to detect both entrepreneurs and non‑entrepreneurs, achieving roughly 60% classification accuracy for each group.

These results highlight how practical experience and interpersonal development boost early‑career salary, while strong academics and a satisfaction‑driven mindset may steer graduates toward entrepreneurial paths.

Actionable Takeaways

Building on these insights:

Students:
Prioritize internships and soft‑skill development to boost early‑career earnings; nurture satisfaction and academic excellence if entrepreneurship is a goal.

Advisors and Universities:
Facilitate practical experiences and communication training, especially for non‑STEM majors, to narrow salary gaps and support entrepreneurial pathways.